[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240125184031.5ddc5799@kernel.org>
Date: Thu, 25 Jan 2024 18:40:31 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Tariq Toukan <ttoukan.linux@...il.com>
Cc: Gal Pressman <gal@...dia.com>, Saeed Mahameed <saeed@...nel.org>, "David
S. Miller" <davem@...emloft.net>, Paolo Abeni <pabeni@...hat.com>, Eric
Dumazet <edumazet@...gle.com>, Saeed Mahameed <saeedm@...dia.com>,
netdev@...r.kernel.org, Tariq Toukan <tariqt@...dia.com>
Subject: Re: [net-next 10/15] net/mlx5e: Let channels be SD-aware
On Thu, 25 Jan 2024 10:01:05 +0200 Tariq Toukan wrote:
> Exactly. That's the desired configuration.
> Our driver has the logic to set it in default.
>
> Here's the default XPS on my setup:
>
> NUMA:
> NUMA node(s): 2
> NUMA node0 CPU(s): 0-11
> NUMA node1 CPU(s): 12-23
>
> PF0 on node0, PF1 on node1.
Okay, good that you took care of the defaults, but having a queue per
CPU thread is quite inefficient. Most sensible users will reconfigure
your NICs and remap IRQs and XPS. Which is fine, but we need to give
them the necessary info to do this right - documentation and preferably
the PCIe dev mapping in the new netlink queue API.
Powered by blists - more mailing lists